Convexity theorems for the gradient map on probability measures